Dendrobium officinale powder release control method

Abstract

The application provides a Dendrobium officinale powder release control method, relates to the technical field of drug quality control, and comprises the following steps: acquiring hyperspectral images of Dendrobium officinale powder under different temperature gradients and correcting; extracting average spectra and preprocessing by using an SNV algorithm; determining polysaccharide content; dividing a calibration set and a test set; extracting characteristic wavelengths by using a CARS algorithm; establishing a PLSR model under different temperature gradients; taking the characteristic wavelengths under all temperature gradients as input, and establishing a SO-PLS model; and determining whether to release or not by evaluating the probability risk of test set samples exceeding a control limit through bootstrap sampling. The application combines temperature control with hyperspectral technology, significantly improves prediction accuracy, introduces a probability risk evaluation mechanism, guarantees quality control, improves detection efficiency, and effectively solves the problems of long time consumption and low efficiency of traditional detection methods.

Description

Technical Field

[0001] This invention relates to the field of drug quality control technology, specifically to a method for controlling the release of Dendrobium officinale powder. Background Technology

[0002] Dendrobium officinale Kimura et Migo, belonging to the genus Dendrobium in the Orchidaceae family, is a precious plant with both medicinal and edible value. The polysaccharides, bibenzyl groups, and flavonoids abundant in Dendrobium officinale endow it with various biological activities such as anti-aging, anti-fatigue, regulation of blood sugar, blood pressure, and blood lipids, and relief of gastric ulcers. Among these, polysaccharides, as the core active ingredient, are not only the main carriers of pharmacological effects but also an important basis for product quality control. In the production process of Dendrobium officinale products, powder, as a key intermediate, must have a polysaccharide content within a preset control range before it can be released for further processing. However, the currently widely used polysaccharide detection method, the phenol-sulfuric acid method, suffers from problems such as cumbersome operation and long processing time, leading to unnecessary stoppages in the production process. Therefore, developing a rapid and non-destructive detection technology to achieve rapid release control of intermediate products has become an urgent need in the industry.

[0003] Near-infrared spectroscopy and near-infrared imaging, with their rapid, safe, and non-destructive characteristics, have played a crucial role in the quality characterization of agricultural products, food, and pharmaceuticals. Temperature, a key influencing factor in near-infrared spectroscopy, causes changes in the intensity, width, and position of absorption peaks. To date, no single theory precisely explains how temperature alters the spectrum. In early studies, temperature was often considered a confounding factor. Researchers have devoted considerable effort to correcting the impact of temperature effects on spectral prediction capabilities. Existing techniques include establishing temperature compensation models by combining spectra measured at different temperatures, or reducing prediction errors by incorporating variables such as scattering effects and temperature into standard models. In fact, if temperature can be precisely controlled, it can also be considered a correction method and a source of information. On the one hand, precise temperature control can greatly improve the repeatability of measurement results; on the other hand, the vibrational states and energy level distributions of molecules differ under different temperature gradients. By fusing spectral fingerprint information from multiple temperature gradients, a more comprehensive picture of the composition and content of mixtures can be obtained.

[0004] It is important to note that near-infrared spectroscopy, as an indirect measurement method, often introduces a significant margin of error between the predicted and actual values ​​of active pharmaceutical ingredients. These errors stem from various factors, including sample characteristics such as particle size, water content, porosity, and the near-infrared response intensity of the analyte's functional groups, as well as instrument and environmental influences such as dark current and ambient temperature and humidity. Accurately predicting certain active ingredients using a model with a limited sample set presents a considerable challenge. In large-scale industrial production, a more practical approach is to establish a relatively accurate model and then, based on the uncertainty of the model's predictions, use probabilistic assessments instead of single predicted values ​​to determine whether a sample carries a risk of exceeding control limits. Samples with a probabilistic risk exceeding a preset value can be verified using precise chemical assays; samples with a probabilistic risk below the preset value can be released directly. This method significantly reduces detection time while ensuring manageable risk.

[0005] Currently, although near-infrared spectroscopy has been applied in the field of medicinal material testing, there is limited research on the detection of polysaccharide content in Dendrobium officinale powder. Most existing detection methods rely on spectral data from a single temperature, lacking utilization of the rich information brought about by temperature changes, and failing to consider the importance of model prediction uncertainties in actual production decisions. Existing technologies suffer from low efficiency and unscientific decision-making in the quality control of Dendrobium officinale intermediate products. Summary of the Invention

[0006] To address the problems existing in the prior art, the purpose of this invention is to provide a method for controlling the release of Dendrobium officinale powder that can effectively utilize information from temperature changes and improve detection efficiency.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for controlling the release of Dendrobium officinale powder, comprising the following steps:

[0008] S1: Obtain hyperspectral images of Dendrobium officinale powder under different temperature gradients and perform black and white plate correction;

[0009] S2: Extract the average spectrum of Dendrobium officinale powder and preprocess the average spectrum using the standard normal variable transformation (SNV) algorithm;

[0010] S3: Determination of polysaccharide content in Dendrobium officinale powder using the phenol-sulfuric acid method;

[0011] S4: Divide the samples into a calibration set and a test set. The polysaccharide content range of the test set is included in the calibration set.

[0012] S5: Use the Competitive Adaptive Reweighted Sampling (CARS) algorithm to extract feature wavelengths related to polysaccharide content from the preprocessed average spectrum;

[0013] S6: Based on the spectral characteristic wavelengths and polysaccharide content data under different temperature gradients, establish partial least squares regression (PLSR) models under different temperature gradients respectively.

[0014] S7: Using the spectral characteristic wavelengths under all temperature gradients as input, establish a sequential orthogonal partial least squares model (SO-PLS);

[0015] S8: Based on the Sequential Orthogonal Partial Least Squares (SO-PLS) model, the bootstrap sampling method is used to predict the test set samples and calculate the probability risk of the polysaccharide content of the samples exceeding the control limit, so as to control the release of Dendrobium officinale powder.

[0016] Further: In step S1, a semiconductor temperature control device is used to perform a gradient temperature change operation on the Dendrobium officinale powder, wherein the temperature gradient range is from 0°C to 80°C.

[0017] Furthermore, the semiconductor temperature control device has both heating and cooling functions, and uses a PID controller to control the temperature. Its operating parameters include target temperature, heating / cooling time, sample holding time, and PID value.

[0018] Further: In step S1, the hyperspectral scanning method is line scanning, the optical mode is diffuse reflection, and the black and white plate correction formula is:

[0019]

[0020] Among them, I cal The hyperspectral image of Dendrobium officinale powder after correction, I raw For the raw hyperspectral image of the powder, I white For the hyperspectral image of the whiteboard, I dark This is a hyperspectral image of a blackboard.

[0021] Further: In step S2, the average spectrum extraction step includes: S21: converting the hyperspectral image into a pseudo-color image; S22: determining the Dendrobium officinale powder region on the pseudo-color image; S23: defining a circle inside the powder region with a threshold equal to or less than the powder region as a mask; S24: averaging the spectra of all pixels within the mask to obtain the average spectrum of Dendrobium officinale powder.

[0022] Furthermore: In step S4, the sample partitioning adopts a sample set partitioning algorithm based on joint xy distance (SPXY).

[0023] Further: In step S6, the steps for establishing the partial least squares regression (PLSR) model under different temperature gradients include:

[0024] S61: Select the search range for the principal components of partial least squares regression;

[0025] S62: The optimal number of principal components is determined by using the three-fold cross-validation method based on the principle of minimizing the root mean square error of validation (RMSECV);

[0026] S63: Establish a partial least squares regression model using the determined optimal number of principal components and the spectral characteristic wavelengths under the corresponding temperature gradient;

[0027] S64: By the coefficient of determination (R²) 2 The prediction performance of the partial least squares regression (PLSR) model under different temperature gradients was evaluated using the root mean square error of prediction (RMSEP) and the root mean square error of prediction (RMSEP).

[0028] Further: In step S7, the steps for establishing the sequential orthogonal partial least squares model (SO-PLS) include:

[0029] S71: Fit the polysaccharide content to the spectral characteristic wavelength data at the first temperature using partial least squares regression;

[0030] S72: Calculate the matrix after orthogonalizing the spectral characteristic wavelength data at the second temperature with the eigenvectors of the first temperature;

[0031] S73: Fit the residuals to the orthogonalized matrix using partial least squares regression;

[0032] S74: Repeat the above steps until all spectral characteristic wavelength data under all temperature gradients are integrated into the model.

[0033] Further: In step S8, the probabilistic risk-based release control steps for Dendrobium officinale powder include:

[0034] S81: Use the bootstrap sampling method to perform multiple random samplings with replacement on the calibration set, and build a new sequential orthogonal partial least squares (SO-PLS) model each time.

[0035] S82: Use all bootstrap models to predict the polysaccharide content for each test set sample to obtain a set of predicted values;

[0036] S83: Calculate the proportion of predicted values ​​exceeding the control limits based on the preset upper and lower control limits as the probability risk;

[0037] S84: Samples with a probability risk below the preset threshold are allowed to be released directly; samples with a probability risk above the preset threshold must be retested using the phenol-sulfuric acid method for confirmation.

[0038] The bootstrap sampling is performed 200 times, and each time 80%-90% of the samples are retained to rebuild the sequential orthogonal partial least squares (SO-PLS) model; the preset threshold is 1%-2%.

[0039] Further: In step S3, the specific steps of the phenol-sulfuric acid method are as follows:

[0040] S31: Accurately weigh Dendrobium officinale powder, place it in an Erlenmeyer flask, add water, and heat under reflux;

[0041] S32: After cooling, transfer to a volumetric flask, dilute to volume with water, shake well, and filter; measure the filtrate, add anhydrous ethanol, and refrigerate to precipitate;

[0042] S33: After centrifugation, discard the supernatant, wash the precipitate with ethanol, and centrifuge again; dissolve the precipitate in hot water and make up to volume to obtain the test solution;

[0043] S34: Measure the sample solution, add phenol solution and sulfuric acid, heat in a water bath; after cooling, measure the absorbance and calculate the polysaccharide content.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] I. This invention overcomes technical bias by transforming temperature from a traditional "interference factor" into an "information source." By acquiring hyperspectral images at different temperature gradients, it captures the differences in the vibrational states and energy level distributions of molecules at different temperatures. These differences are manifested spectrally as variations in the intensity, position, and width of functional group absorption peaks. The SO-PLS model effectively integrates spectral feature information from different temperatures, extracting incremental information to achieve a more comprehensive description of the material composition and significantly improve the accuracy of the polysaccharide content prediction model.

[0046] Second, the traditional phenol-sulfuric acid method for determining polysaccharide content is cumbersome and time-consuming, while the near-infrared hyperspectral detection method of this invention can complete data acquisition and analysis within minutes, significantly shortening the detection time. More importantly, the probabilistic risk assessment mechanism introduced in this invention can directly determine whether a sample should be released based on the uncertainty of the predicted results. For samples with low risk, there is no need for further chemical determination, further improving production efficiency and reducing unnecessary stoppages in the production process.

[0047] Third, this invention overcomes the limitations of traditional single-prediction-value judgment. It assesses the uncertainty of model predictions through a bootstrap sampling method, generating a probability distribution of polysaccharide content for each test sample, and then quantitatively evaluates the risk probability of exceeding control limits. This maximizes production efficiency while ensuring product quality, balancing the relationship between testing efficiency and quality control.

[0048] Fourth, the technical solution proposed in this invention is not only applicable to the detection of polysaccharide content in Dendrobium officinale powder, but can also be extended to the determination and quality control of active ingredient content in other Chinese medicinal materials and natural products. Especially in pharmaceutical processes where strict control of intermediate product quality consistency is required, this method can effectively improve lean manufacturing levels and has broad application prospects. Attached Figure Description

[0049] Figure 1 This is a flowchart of a method for controlling the release of Dendrobium officinale powder based on temperature-controlled hyperspectral imaging according to the present invention.

[0050] Figure 2 This is a schematic diagram of a temperature-controlled near-infrared hyperspectral imaging device according to an embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram of a pseudo-color image generated at wavelengths of 1598.92nm, 1398.65nm, and 1198.49nm, according to an embodiment of the present invention.

[0052] Figure 4 This is a schematic diagram of a mask image on a pseudo-color image generated at wavelengths of 1598.92nm, 1398.65nm, and 1198.49nm, according to an embodiment of the present invention.

[0053] Figure 5 This is a schematic diagram of the average spectrum of a powder according to an embodiment of the present invention;

[0054] Figure 6 (A) in the figure is a schematic diagram of the average spectrum curve after SNV preprocessing according to an embodiment of the present invention; Figure 6 (B) is a partially enlarged schematic diagram of the average spectrum after SNV preprocessing according to an embodiment of the present invention;

[0055] Figure 7 (A) is a schematic diagram of the probability distribution formed by the average value, control upper and lower limits, and bootstrap sampling of the test set samples numbered 1-9 in an embodiment of the present invention. Figure 7 (B) is a schematic diagram of the probability distribution formed by the average value, control upper and lower limits, and bootstrap sampling of the test set samples numbered 10-18 in an embodiment of the present invention. Figure 7 (C) in the figure is a schematic diagram of the probability distribution formed by the average value, control upper and lower limits and bootstrap sampling of the test set samples numbered 19-27 in an embodiment of the present invention. Figure 7(D) in the figure is a schematic diagram of the probability distribution formed by the average value, control upper and lower limits and bootstrap sampling of the test set samples numbered 28-35 in an embodiment of the present invention. Detailed Implementation

[0056] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] It should be noted that in the description of this invention, SNV is the abbreviation for Standard Normal Variable Transformation Algorithm, and has the same meaning as the abbreviation for Standard Normal Variable Transformation Algorithm. Similarly, CARS is the abbreviation for Competitive Adaptive Reweighted Sampling Algorithm, PLSR is the abbreviation for Partial Least Squares Regression Model, SO-PLS is the abbreviation for Sequential Orthogonal Partial Least Squares Model, SPXY is the abbreviation for Sample Set Partitioning Algorithm, RMSECV is the abbreviation for Root Mean Square Error of Verification, and R... 2 The English abbreviation for the coefficient of determination is RMSEP (Root Mean Square Error of Prediction).

[0058] A method for controlling the release of Dendrobium officinale powder includes the following steps:

[0059] S1: Obtain hyperspectral images of Dendrobium officinale powder under different temperature gradients and perform black and white plate correction;

[0060] S2: Extract the average spectrum of Dendrobium officinale powder and preprocess the average spectrum using the standard normal variable transformation (SNV) algorithm;

[0061] S3: Determination of polysaccharide content in Dendrobium officinale powder using the phenol-sulfuric acid method;

[0062] S4: Divide the samples into a calibration set and a test set. The polysaccharide content range of the test set is included in the calibration set.

[0063] S5: Use the Competitive Adaptive Reweighted Sampling (CARS) algorithm to extract feature wavelengths related to polysaccharide content from the preprocessed average spectrum;

[0064] S6: Based on the spectral characteristic wavelengths and polysaccharide content data under different temperature gradients, establish partial least squares regression (PLSR) models under different temperature gradients respectively.

[0065] S7: Using the spectral characteristic wavelengths under all temperature gradients as input, establish a sequential orthogonal partial least squares model (SO-PLS);

[0066] S8: Based on the Sequential Orthogonal Partial Least Squares (SO-PLS) model, the bootstrap sampling method is used to predict the test set samples and calculate the probability risk of the polysaccharide content of the samples exceeding the control limit, so as to control the release of Dendrobium officinale powder.

[0067] Further: In step S1, a semiconductor temperature control device is used to perform a gradient temperature change operation on the Dendrobium officinale powder, wherein the temperature gradient range is from 0°C to 80°C.

[0068] Furthermore, the semiconductor temperature control device has both heating and cooling functions, and uses a PID controller to control the temperature. Its operating parameters include target temperature, heating / cooling time, sample holding time, and PID value.

[0069] Further: In step S1, the hyperspectral scanning method is line scanning, the optical mode is diffuse reflection, and the black and white plate correction formula is:

[0070]

[0071] Among them, I cal For the corrected hyperspectral image of Dendrobium officinale powder, I raw For the raw hyperspectral image of the powder, I white For the hyperspectral image of the whiteboard, I dark This is a hyperspectral image of a blackboard.

[0072] Further: In step S2, the average spectrum extraction step includes: S21: converting the hyperspectral image into a pseudo-color image; S22: determining the Dendrobium officinale powder region on the pseudo-color image; S23: defining a circle inside the powder region with a threshold equal to or less than the powder region as a mask; S24: averaging the spectra of all pixels within the mask to obtain the average spectrum of Dendrobium officinale powder.

[0073] Furthermore: In step S4, the sample partitioning adopts a sample set partitioning algorithm based on joint xy distance (SPXY).

[0074] Further: In step S6, the steps for establishing the partial least squares regression (PLSR) model under different temperature gradients include:

[0075] S61: Select the search range for the principal components of partial least squares regression;

[0076] S62: The optimal number of principal components is determined by using the three-fold cross-validation method based on the principle of minimizing the root mean square error of validation (RMSECV);

[0077] S63: Establish a partial least squares regression model using the determined optimal number of principal components and the spectral characteristic wavelengths under the corresponding temperature gradient;

[0078] S64: By the coefficient of determination (R²) 2 The prediction performance of the partial least squares regression (PLSR) model under different temperature gradients was evaluated using the root mean square error of prediction (RMSEP) and the root mean square error of prediction (RMSEP).

[0079] Further: In step S7, the steps for establishing the sequential orthogonal partial least squares model (SO-PLS) include:

[0080] S71: Fit the polysaccharide content to the spectral characteristic wavelength data at the first temperature using partial least squares regression;

[0081] S72: Calculate the matrix after orthogonalizing the spectral characteristic wavelength data at the second temperature with the eigenvectors of the first temperature;

[0082] S73: Fit the residuals to the orthogonalized matrix using partial least squares regression;

[0083] S74: Repeat the above steps until all spectral characteristic wavelength data under all temperature gradients are integrated into the model.

[0084] Further: In step S8, the probabilistic risk-based release control steps for Dendrobium officinale powder include:

[0085] S81: Use the bootstrap sampling method to perform multiple random samplings with replacement on the calibration set, and build a new sequential orthogonal partial least squares (SO-PLS) model each time.

[0086] S82: Use all bootstrap models to predict the polysaccharide content for each test set sample to obtain a set of predicted values;

[0087] S83: Calculate the proportion of predicted values ​​exceeding the control limits based on the preset upper and lower control limits as the probability risk;

[0088] S84: Samples with a probability risk below the preset threshold are allowed to be released directly; samples with a probability risk above the preset threshold must be retested using the phenol-sulfuric acid method for confirmation.

[0089] The bootstrap sampling is performed 200 times, and each time 80%-90% of the samples are retained to rebuild the sequential orthogonal partial least squares (SO-PLS) model; the preset threshold is 1%-2%.

[0090] Further: In step S3, the specific steps of the phenol-sulfuric acid method are as follows:

[0091] S31: Accurately weigh the Dendrobium officinale powder, place it in an Erlenmeyer flask, add water, and heat under reflux;

[0092] S32: After cooling, transfer to a volumetric flask, dilute to volume with water, shake well, and filter; measure the filtrate, add anhydrous ethanol, and refrigerate to precipitate;

[0093] S33: After centrifugation, discard the supernatant, wash the precipitate with ethanol, and centrifuge again; dissolve the precipitate in hot water and make up to volume to obtain the test solution;

[0094] S34: Measure the sample solution, add phenol solution and sulfuric acid, heat in a water bath; after cooling, measure the absorbance and calculate the polysaccharide content.

[0095] The implementation process and effects of the method of the present invention will be described in detail below through specific embodiments.

[0096] Example 1

[0097] like Figure 1 The following is an example of a method for controlling the release of Dendrobium officinale powder according to one embodiment of the present invention, comprising the following steps:

[0098] S1: Obtain hyperspectral images of Dendrobium officinale powder under different temperature gradients and perform black and white plate correction.

[0099] Specifically, a sufficient amount of Dendrobium officinale powder was collected and ground into a 300-mesh fine powder. A series of temperature sampling points were set within a temperature range of 0℃ to 80℃, and a semiconductor temperature control device was used to perform gradient temperature control on the powder. After reaching a certain temperature sampling point, the sample was kept at that temperature for a period of time, its hyperspectral image was scanned, and then the hyperspectral image was corrected using a black and white plate.

[0100] Furthermore, the semiconductor temperature control device should simultaneously possess heating and cooling functions, using a PID controller to control the temperature. Its operating parameters include target temperature, heating / cooling time, sample holding time, and PID values ​​(P, I, and D values). To ensure detection efficiency, the heating / cooling time should not exceed 2 minutes, and the sample holding time should not exceed 5 minutes. The hyperspectral scanning method is line scanning, the optical mode is diffuse reflection, and a stepper motor drives the sample through the lens. Its operating parameters include exposure time, stepper motor speed, start scanning position, end scanning position, distance between lens and sample, imaging resolution, maximum reflection intensity, spectral range, spectral resolution, and number of wavelengths. The blackboard used for hyperspectral image correction is a black PTFE lens cap, and the whiteboard is a white PTFE plate with a diffuse reflection surface. The black-and-white correction formula is:

[0101]

[0102] Among them, I cal For the powder-corrected hyperspectral image, Iraw For the raw hyperspectral image of the powder, I white For the hyperspectral image of the whiteboard, I dark This is a hyperspectral image of a blackboard.

[0103] S2: Extract the average spectrum of Dendrobium officinale powder and preprocess the average spectrum using the standard normal variable transformation (SNV) algorithm.

[0104] Specifically, to extract the average spectrum of the powder region, the hyperspectral image needs to be converted into a pseudo-color image first. The grayscale images of the R (red), G (green), and B (blue) channels of the pseudo-color image are defined as the grayscale images at 1598.92 nm, 1398.65 nm, and 1198.49 nm of the hyperspectral image, respectively. Then, the powder region is visually identified on the pseudo-color image, and a circle slightly smaller than the powder region is defined as a mask. The average spectrum of the powder is obtained by averaging the spectra of all pixels within the mask.

[0105] The SNV algorithm is mainly used to eliminate the influence of solid particle size, surface scattering, and optical path variation on near-infrared diffuse reflectance spectra. The formula for SNV transformation is:

[0106]

[0107] Where x is the average spectrum, x SNV The average spectrum after SNV processing. m is the number of wavelengths in the average spectrum, i = 1, 2, ..., m.

[0108] S3: The polysaccharide content of Dendrobium officinale powder was determined using the phenol-sulfuric acid method.

[0109] Specifically, the phenol-sulfuric acid method is a common method for determining polysaccharide content. It is based on the fact that polysaccharides hydrolyze to monosaccharides under strong acid conditions, and the monosaccharides react with phenol to form colored compounds. The absorbance of these compounds is measured by colorimetry, thereby estimating the polysaccharide content.

[0110] The specific steps are as follows:

[0111] S31: Accurately weigh about 0.3g of Dendrobium officinale powder, place it in a 250mL Erlenmeyer flask, add 200mL of water, and heat under reflux for 2 hours using a hot plate.

[0112] S32: After cooling, transfer to a 250mL volumetric flask. Wash the conical flask three times with a small amount of water, and combine the washings with the same volumetric flask. Add water to the mark, shake well, and filter through a dry filter. Accurately measure 2mL of the filtrate and place it in a 15mL centrifuge tube. Accurately add 10mL of anhydrous ethanol, shake well, and refrigerate at 4℃ for 1 hour.

[0113] S33: Remove the sample, centrifuge at 4000 rpm for 20 min, discard the supernatant, wash the precipitate twice with 8 mL of 80% ethanol each time. Centrifuge at 4000 rpm for 20 min, discard the supernatant, dissolve the precipitate in hot water, transfer to a 25 mL volumetric flask, cool, add water to the mark, and shake well to obtain the test solution.

[0114] S34: Accurately measure 1 mL of the test solution and place it in a 10 mL stoppered test tube. Quickly and accurately add 1 mL of 5% phenol solution, shake well, and then accurately add 5 mL of sulfuric acid along the wall of the test tube. Shake well again and heat in a boiling water bath for 20 min. After removal, cool in an ice-water bath for 5 min and measure the absorbance at 488 nm. Repeat the measurement three times for each sample. The formula for calculating the polysaccharide content of Dendrobium officinale powder is:

[0115]

[0116] Where c is the polysaccharide concentration of the test solution, and M is the weight of the accurately weighed Dendrobium officinale powder.

[0117] S4: Use the sample set partitioning algorithm based on joint xy distance (SPXY) to partition the samples into a calibration set and a test set.

[0118] Specifically, when calculating the distance between samples, the SPXY algorithm considers not only the feature dimension direction (x vector) but also the true value dimension direction (y vector), which can increase the difference and representativeness between samples.

[0119] S5: Use the Competitive Adaptive Reweighted Sampling (CARS) algorithm to extract feature wavelengths related to polysaccharide content from the preprocessed average spectrum.

[0120] Specifically, the CARS algorithm, by simulating biological evolution, adaptively competes for and reweights spectral wavelengths to efficiently select the most useful features for the model, making it suitable for feature selection tasks involving high-dimensional data. Its core lies in iterative optimization, gradually eliminating relatively unimportant features to improve the model's predictive performance and interpretability. The CARS algorithm requires setting three hyperparameters: the number of principal components, the number of cross-validation folds, and the number of iterations. Its specific process is as follows:

[0121] S51: Using the Monte Carlo sampling method, a certain proportion (usually 80%) of the samples are selected from the calibration set for modeling each time, and the remaining samples are used as the validation set for validation to establish the PLS model. The number of Monte Carlo samplings, N, is set, and the absolute value weights of the regression coefficients of the PLS model are calculated for each sampling process. The calculation formula is:

[0122]

[0123] Among them, |b i| represents the absolute value of the regression coefficient for the i-th wavelength of the average spectrum after SNV processing, w i is the absolute value weight of the regression coefficient of the i-th wavelength of the average spectrum after SNV processing, and p is the number of wavelengths remaining after each sampling.

[0124] S52: Remove w using the Exponentially Decreasing Function (EDF). i Relatively small wavelengths. After the j-th Monte Carlo sampling, the wavelength proportion R retained by the EDF is... j for:

[0125] R j =μe -kj

[0126] Where μ and k are two constants, P represents the total number of wavelengths in the original spectrum.

[0127] S53: During each sampling, a specific number of wavelengths are selected from the previous sampling using Adaptive Reweighted Sampling (ARS) to build a PLS model, and the root mean square error of validation (RMSEV) is calculated on the validation set.

[0128] S54: After N samplings, N candidate feature wavelength subsets and their corresponding RMSEV values ​​are obtained. The feature wavelength subset corresponding to the minimum RMSEV value is selected as the final result of wavelength screening.

[0129] S6: Based on the spectral characteristic wavelengths and polysaccharide content data under different temperature gradients, partial least squares regression (PLSR) models under different temperature gradients were established respectively.

[0130] Specifically, hyperspectral images under different temperature gradients were acquired. After black-and-white plate correction, average spectrum extraction and preprocessing, and feature wavelength extraction, spectral feature wavelengths under different temperature gradients were obtained and used to build PLS models under different temperature gradients. The PLS models used a three-fold cross-validation method to optimize hyperparameters, with the principal component PC1 as the hyperparameter. Root mean square error (RMSE) and the coefficient of determination (R²) were used. 2 To evaluate the accuracy of the PLS model.

[0131] Triple-fold cross-validation further divides the calibration set into a training set and a cross-validation set. The RMSE of the training set, cross-validation set, and test set are denoted as RMSEC, RMSECV, and RMSEP, respectively, and their R... 2 Represented as R c 2 R cv 2 and R p 2 The hyperparameters of the model are determined according to the principle of minimizing RMSECV, and then the model is fitted using data. A better model should have a relatively high R-value. p 2 And a relatively low RMSEP.

[0132] S7: Using the spectral characteristic wavelengths under all temperature gradients as input, establish a sequential orthogonal partial least squares (SO-PLS) model.

[0133] Specifically, the SO-PLS model is established using the spectral characteristic wavelengths under all temperature gradients as input. The SO-PLS model uses three-fold cross-validation to optimize hyperparameters, with the principal component count (PC2) as the hyperparameter. RMSE and R... 2 The accuracy of the SO-PLS model is evaluated using the same method as that used for the PLS model described above.

[0134] SO-PLS extracts incremental information from spectral characteristic wavelength data under multiple temperature gradients, thereby gradually improving the model's predictive performance. This can be summarized as follows:

[0135] S71: Assume X1 and X2 represent two sets of preprocessed spectral data, and Y is their common response. Y is fitted to X1 using PLS regression:

[0136]

[0137] in, For feature vectors, e1 is the load, and e1 is the residual.

[0138] S72: Calculate X2 and T X1 Orthogonal parts:

[0139]

[0140] S73:e1 via PLS regression and Perform fitting:

[0141]

[0142] S74: Y is regressed using PLS regression and and Perform fitting:

[0143]

[0144] If more than two sets of spectral characteristic wavelength data are involved in the modeling, repeat steps S72 to S74.

[0145] It should be noted that under different temperature gradients, the vibrational states and energy level distributions of molecules will differ, which will manifest in the spectrum as changes in the intensity, position, and width of functional group absorption peaks. The SO-PLS model can more effectively capture the subtle differences between spectral signals measured under different temperature conditions, extract the incremental information, and thus more comprehensively depict the material composition and content information of the mixture, which is beneficial to improving the accuracy of polysaccharide content prediction models.

[0146] S8: Based on the SO-PLS model, the bootstrap sampling method is used to predict the test set samples and calculate the probability risk of the polysaccharide content of the samples exceeding the control limit, so as to control the release of Dendrobium officinale powder.

[0147] Specifically, the bootstrap sampling method is used to perform 200 random samplings with replacement on the calibration set, retaining 80%-90% of the samples each time to rebuild the SO-PLS model. After each SO-PLS model training, the polysaccharide content of the test set is re-predicted, resulting in 200 predicted content values ​​for each sample in the test set. Then, based on the set upper and lower control limits, the probabilistic risk is calculated to determine whether the powder sample is allowed to proceed to the next process step.

[0148] Furthermore, the probabilistic risk is calculated independently for each sample in the test set. The set of 200 predicted content values ​​is denoted as Pred, and the upper and lower control limits are denoted as L. upper and L lower If all elements in set Pred are in L lower and L upper Between these values, the probability risk RI of the sample is 0. If there exists an element less than L in the set Pred... lower or greater than L upper The number of elements exceeding the upper and lower limits (LN) is counted, and the probability risk of the sample is LN / 200. Samples with a probability risk below 2% can proceed to the next process step. Otherwise, the polysaccharide content should be re-determined using the phenol-sulfuric acid method.

[0149] It should be noted that this invention overcomes the technical bias that temperature is a confounding factor in near-infrared spectroscopy. Temperature, as a key influencing factor in near-infrared spectroscopy, causes changes in the intensity, width, and position of absorption peaks. To date, no theory can precisely explain how temperature changes the spectrum. In early related studies, temperature was often treated as a confounding factor. Researchers have made considerable efforts to correct the impact of temperature effects on spectral prediction capabilities. For example, S. Kawano et al. established a temperature compensation model by combining spectra measured at different temperatures. M. Tarumi et al. reduced prediction errors by incorporating variables such as scattering effects and temperature into the standard model. This invention employs a precise temperature control technique. On the one hand, precise temperature control greatly improves the repeatability of measurement results; on the other hand, the vibrational states and energy level distributions of molecules differ under different temperature gradients. By fusing spectral fingerprint information from multiple temperature gradients, a more comprehensive description of the composition and content information of mixtures can be obtained.

[0150] This invention employs a quantitative calibration model establishment method between spectral characteristic wavelengths and polysaccharide content under a full temperature gradient. The temperature of *Dendrobium officinale* powder is precisely adjusted using a semiconductor temperature control device, followed by hyperspectral scanning to obtain a series of hyperspectral images of the powder under different temperature gradients. After black-and-white plate calibration, a mask region is determined within the powder area, and the average spectral signal is extracted. After SNV preprocessing and CARS characteristic wavelength extraction, the average spectrum is converted into spectral characteristic wavelengths. Using the spectral characteristic wavelengths under different temperature gradients as input and the polysaccharide content as output, an SO-PLS model is established. Under different temperature gradients, the vibrational states and energy level distributions of molecules differ, manifesting as changes in the intensity, position, and width of functional group absorption peaks in the spectrum. The SO-PLS model can more effectively capture the subtle differences between spectral signals measured under different temperature conditions, extracting incremental information and thus more comprehensively depicting the material composition and content information of the mixture, which is beneficial for improving the accuracy of the polysaccharide content prediction model.

[0151] This invention uses the bootstrap algorithm to measure the uncertainty of the SO-PLS model, generating a set containing multiple predicted polysaccharide content values ​​through repeated sampling. The probability risk of the current sample is assessed by counting the number of elements in this set that exceed the control limits. When the probability risk is less than a preset value, the sample can be released to the next process stage; otherwise, precise chemical analysis is required to reconfirm whether its polysaccharide content exceeds the limit. This significantly improves detection efficiency and reduces detection costs while ensuring the consistency of intermediate products in the pharmaceutical process, effectively enhancing the lean manufacturing level of the Dendrobium officinale industry.

[0152] This invention precisely regulates the temperature of Dendrobium officinale powder using a semiconductor temperature control device, then performs hyperspectral scanning to obtain a series of hyperspectral images of the powder under different temperature gradients. After black-and-white plate correction, a mask region is determined in the powder area, and the average spectral signal is extracted. After SNV preprocessing and CARS characteristic wavelength extraction, the average spectrum is converted into spectral characteristic wavelengths. Using the spectral characteristic wavelengths under different temperature gradients as input and polysaccharide content as output, an SO-PLS model is established. Under different temperature gradients, the vibrational states and energy level distributions of molecules differ, which are reflected in the spectrum as changes in the intensity, position, and width of functional group absorption peaks. The SO-PLS model can more effectively capture the subtle differences between spectral signals measured under different temperature conditions, extract incremental information, and thus more comprehensively depict the material composition and content information of the mixture, which is beneficial to improving the accuracy of the polysaccharide content prediction model.

[0153] It should be noted that the spectral detection method in this invention can not only use a near-infrared hyperspectral imaging system to acquire near-infrared spectral signals, but also a Fourier transform near-infrared spectrometer and a Fabry-Perot near-infrared spectrometer to acquire near-infrared spectral signals.

[0154] It should be noted that the SO-PLS algorithm in this invention can also be replaced by the following algorithm: Multi-Block Partial Least Squares (MB-PLS): The MB-PLS algorithm is a statistical method for processing the relationships between multiple sets of data blocks. MB-PLS can extract principal components from data blocks from different sources for comprehensive analysis, establish a relationship model between them, and explain the potential connections between the variables represented by different data blocks.

[0155] Example 2

[0156] The algorithm in this embodiment was compiled using Python (v3.9.12; Python Software Foundation, 2022) and MATLAB R2018b. It includes the following steps:

[0157] S1: Collect 70 batches of Dendrobium officinale samples, grind them into powder, and pass them through a 300-mesh sieve to obtain the test samples. Temperature-controlled near-infrared hyperspectral imaging device, such as... Figure 2 As shown, the temperature gradient points are set to 10℃ and 50℃, and a semiconductor temperature control device (provided by OptoMeasuringFuture (Shenzhen) Technology Co., Ltd.) is used to perform the gradient temperature operation.

[0158] The sample was packed and compacted in a pure copper mold with a hole depth of 5 mm and a hole diameter of 19.05 mm. Each sample was first heated to 10℃ and then heated to 50℃. The semiconductor temperature control device has both heating and cooling functions, using a PID controller to control the temperature. The operating settings were: heating / cooling time 1.5 minutes, sample holding time 3 minutes, P value 3000, I value 150, D value 0. The hyperspectral scanning method was line scanning, the optical mode was diffuse reflection, and a stepper motor moved the sample through the lens. The operating parameters were: exposure time 35 ms, stepper motor speed 1.13 mm / s, start scanning position 150 mm, end scanning position 210 mm, distance between lens and sample 16.5 cm, imaging resolution 638×512, maximum reflectance intensity approximately 11500, spectral range 898.47 nm~1750.87 nm, spectral resolution 1.7 nm, and number of wavelengths 512.

[0159] Because the grayscale image at wavelength 898.47nm contained bad pixels, and the spectral curves between wavelengths 1692.42nm and 1750.87nm contained significant noise, these wavelengths were discarded. Ultimately, 474 wavelengths from the hyperspectral image between 900.14nm and 1690.75nm were retained for subsequent analysis. After performing black-and-white correction, the spectra of all pixels in the hyperspectral image were converted from reflectance intensity to reflectance, with reflectance ranging from 0 to 1.

[0160] S2: As Figures 3-4 As shown, grayscale images at 1598.92nm, 1398.65nm, and 1198.49nm of the hyperspectral image were selected as the R, G, and B channels, respectively, to generate pseudo-color images. Due to errors in the stepper motor's travel, the image resolution of the pseudo-color images ranged from 786×638 to 829×638. A circle was drawn with the lower left corner of the pseudo-color image as the origin, the pixel coordinates (340, 430) as the center, and 70 pixels as the radius. The area within this circle was used as a mask. Figure 5 As shown, the average spectrum of the powder is obtained by averaging the spectra of all pixels within the mask, as shown below. Figure 6 As shown, the average spectrum is obtained by preprocessing the average spectrum using the SNV algorithm. Figure 6 (A) in the figure is a schematic diagram of the average spectrum curve after SNV preprocessing according to an embodiment of the present invention; Figure 6 Image (B) is a partially enlarged schematic diagram of the average spectrum after SNV preprocessing according to an embodiment of the present invention. Figures 5-6 The solid line represents the spectral curve at 10℃, and the dotted line represents the spectral curve at 50℃.

[0161] S3: The polysaccharide content was determined using the phenol-sulfuric acid method. The polysaccharide content of 70 batches of samples ranged from 25.80% to 64.51%, with an average of 51.57% and a standard deviation of 6.62%.

[0162] S4: Using the SPXY algorithm, all samples were divided into a calibration set and a test set in a 1:1 ratio. After the division, the polysaccharide content of the calibration set ranged from 25.80% to 64.51%, with an average of 50.65% and a standard deviation of 7.67%; the polysaccharide content of the test set ranged from 39.50% to 64.24%, with an average of 52.48% and a standard deviation of 5.20%. The polysaccharide content range of the test set was included within that of the calibration set, ensuring that the trained model had good generalization ability.

[0163] S5: The principal component count of the CARS algorithm was set to 15, the cross-validation fold count to 3, and the number of iterations to 10,000. The CARS algorithm was used to extract characteristic wavelengths from the preprocessed spectra at 10℃ and 50℃. At 10℃, 20 characteristic wavelengths were extracted, including 1086.79nm, 1088.46nm, and 1116.80nm. At 50℃, 33 characteristic wavelengths were extracted, including 900.14nm, 938.46nm, and 940.13nm.

[0164] S6: A PLS model was established based on the spectral characteristic wavelength at 10℃. The range of principal components at 10℃ was set to 1–30. The number of principal components was optimized using a grid search method, and the optimal principal component for three-fold cross-validation was determined to be 8. A PLS model was then established based on the spectral characteristic wavelength at 50℃. The range of principal components at 50℃ was set to 1–30, and the optimal principal component for three-fold cross-validation was determined to be 13. On the test set, the 50℃ PLS model achieved better results than the 10℃ PLS model, R... 2 p reached 0.850, and RMSEP reached 2.01%.

[0165] Table 1. Quantitative results of the PLS model at 10℃ and 50℃

[0166]

[0167] S7: Using the spectral characteristic wavelengths obtained at 10℃ and 50℃ as input, an SO-PLS model was established. During model construction, the hyperparameters were optimized using a grid method, and the search range for the number of principal components was set to 1–30. After three-fold cross-validation, the optimal number of principal components was determined to be 18. The prediction performance of the SO-PLS model on the test set was improved compared to the PLS model at 50℃, R0. 2 The p-value increased by 0.045 and the RMSEP decreased by approximately 0.33%, indicating that the model has a certain predictive ability.

[0168] Table 2 Quantitative Results of SO-PLS Model

[0169]

[0170] S8: Use the bootstrap sampling method to perform 200 random samplings with replacement on the calibration set, retaining 80%-90% of the samples each time to rebuild the SO-PLS model. After each SO-PLS model training, re-predict the polysaccharide content of the test set. Ultimately, each sample in the test set corresponds to 200 predicted content values, forming a probability distribution. Based on the polysaccharide content distribution range of all samples, set upper and lower control limits: the upper control limit is set to the mean + 1.5 × standard deviation, and the lower control limit is set to the mean - 1.5 × standard deviation, which are 61.50% and 41.64%, respectively. Figure 7 As shown, a schematic diagram of the probability distribution of 35 test set samples formed by the mean, control limits, and bootstrap sampling is presented. Figure 7 (A) is a schematic diagram of the probability distribution formed by the average value, control upper and lower limits, and bootstrap sampling of the test set samples numbered 1-9 in an embodiment of the present invention. Figure 7 (B) is a schematic diagram of the probability distribution formed by the average value, control upper and lower limits, and bootstrap sampling of the test set samples numbered 10-18 in an embodiment of the present invention. Figure 7 (C) in the figure is a schematic diagram of the probability distribution formed by the average value, control upper and lower limits and bootstrap sampling of the test set samples numbered 19-27 in an embodiment of the present invention. Figure 7 (D) in the diagram is a schematic diagram of the probability distribution formed by the average value, control upper and lower limits, and bootstrap sampling of test set samples numbered 28-35 according to an embodiment of the present invention. The average value is within... Figure 7 The upper limit is represented by a dashed line, the lower limit by a dotted line, and the middle limit by a dotted line.

[0171] Taking the samples from test sets 1, 2, 3, and 4 as examples, the first sample was 100% within the control range with a probability risk of 0, and was directly released to the next process stage; the second sample had a 27% probability of being within the control range with a probability risk of 73%, requiring re-precision chemical determination, and its actual polysaccharide content was highly likely to exceed the control limit; the third sample had a 93% probability of being within the control range with a probability risk of 7%, which was higher than the preset standard of 2%, and also required re-precision chemical determination; the fourth sample had a 99.5% probability of being within the control range with a probability risk of 0.5%, which was lower than the preset standard of 2%, and was directly released to the next process stage.

[0172] As can be seen from the above embodiments, the release control method for Dendrobium officinale powder based on temperature-controlled near-infrared hyperspectral analysis and probabilistic risk assessment proposed in this invention successfully solves the problems of low efficiency and unscientific decision-making in traditional methods by introducing temperature control, multi-temperature spectral information fusion, and probabilistic risk assessment. It not only improves the accuracy of polysaccharide content prediction but also significantly improves detection efficiency, while achieving scientific quality risk control, providing a new technical path for the production quality control of Dendrobium officinale and similar natural medicinal materials.

[0173] The above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. They should not be construed as limiting the scope of protection of the present invention. All equivalent transformations or modifications made in accordance with the spirit and essence of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for controlling the release of Dendrobium officinale powder, characterized in that, Includes the following steps: S1: Obtain hyperspectral images of Dendrobium officinale powder under different temperature gradients and perform black and white plate correction; S2: Extract the average spectrum of Dendrobium officinale powder and preprocess the average spectrum using a normal variable transformation algorithm; S3: Determination of polysaccharide content in Dendrobium officinale powder using the phenol-sulfuric acid method; S4: Divide the samples into a calibration set and a test set. The polysaccharide content range of the test set is included in the calibration set. S5: Use a competitive adaptive reweighted sampling algorithm to extract feature wavelengths related to polysaccharide content from the preprocessed average spectrum; S6: Based on the spectral characteristic wavelengths and polysaccharide content data under different temperature gradients, establish partial least squares regression models under different temperature gradients respectively; S7: Using the spectral characteristic wavelengths under all temperature gradients as input, establish a sequential orthogonal partial least squares model; S8: Based on the sequential orthogonal partial least squares model, the bootstrap sampling method is used to predict the test set samples and calculate the probability risk of the polysaccharide content of the samples exceeding the control limit, so as to control the release of Dendrobium officinale powder.

2. The method for controlling the release of Dendrobium officinale powder according to claim 1, characterized in that: In step S1, a semiconductor temperature control device is used to perform a gradient temperature change operation on the Dendrobium officinale powder, wherein the temperature gradient range is from 0°C to 80°C.

3. The method for controlling the release of Dendrobium officinale powder according to claim 2, characterized in that: The semiconductor temperature control device has both heating and cooling functions, and uses a PID controller to control the temperature. Its operating parameters include target temperature, heating time, cooling time, sample holding time, and PID value.

4. The method for controlling the release of Dendrobium officinale powder according to claim 1, characterized in that: In step S1, the hyperspectral scanning method is line scanning, the optical mode is diffuse reflection, and the black and white plate correction formula is: Among them, I cal For the corrected hyperspectral image of Dendrobium officinale powder, I raw For the raw hyperspectral image of the powder, I white For the hyperspectral image of the whiteboard, I dark This is a hyperspectral image of a blackboard.

5. The method for controlling the release of Dendrobium officinale powder according to claim 1, characterized in that: In step S2, the average spectrum extraction step includes: S21: converting the hyperspectral image into a pseudo-color image; S22: determining the Dendrobium officinale powder region on the pseudo-color image; S23: defining a circle inside the powder region with a threshold equal to or less than the powder region as a mask; S24: averaging the spectra of all pixels within the mask to obtain the average spectrum of Dendrobium officinale powder.

6. The method for controlling the release of Dendrobium officinale powder according to claim 1, characterized in that: In step S4, the sample partitioning adopts a sample set partitioning algorithm based on joint xy distance.

7. The method for controlling the release of Dendrobium officinale powder according to claim 1, characterized in that: In step S6, the establishment of the partial least squares regression model under different temperature gradients includes the following steps: S61: Select the search range for the principal components of partial least squares regression; S62: The optimal number of principal components is determined by using the three-fold cross-validation method based on the principle of minimizing the root mean square error of validation. S63: Establish a partial least squares regression model using the determined optimal number of principal components and the spectral characteristic wavelengths under the corresponding temperature gradient; S64: Evaluate the predictive performance of the partial least squares regression model under different temperature gradients by using the coefficient of determination and the root mean square error of prediction.

8. The method for controlling the release of Dendrobium officinale powder according to claim 1, characterized in that: In step S7, the steps for establishing the sequential orthogonal partial least squares model include: S71: Fit the polysaccharide content to the spectral characteristic wavelength data at the first temperature using partial least squares regression; S72: Calculate the matrix after orthogonalizing the spectral characteristic wavelength data at the second temperature with the eigenvectors of the first temperature; S73: Fit the residuals to the orthogonalized matrix using partial least squares regression; S74: Repeat the above steps until all spectral characteristic wavelength data under all temperature gradients are integrated into the model.

9. The method for controlling the release of Dendrobium officinale powder according to claim 1, characterized in that: In step S8, the probabilistic risk-based release control steps for Dendrobium officinale powder include: S81: Use the bootstrap sampling method to perform multiple random samplings with replacement on the calibration set, and establish a new sequential orthogonal partial least squares model each time. S82: Use all bootstrap models to predict the polysaccharide content for each test set sample to obtain a set of predicted values; S83: Calculate the proportion of predicted values ​​exceeding the control limits based on the preset upper and lower control limits as the probability risk; S84: Samples with a probability risk below the preset threshold are allowed to be released directly; samples with a probability risk higher than or equal to the preset threshold must be retested using the phenol-sulfuric acid method for confirmation. The bootstrap sampling is performed 200 times, and each time 80%-90% of the samples are retained to rebuild the sequential orthogonal partial least squares model; the preset threshold is 1%-2%.

10. A method for controlling the release of Dendrobium officinale powder according to any one of claims 1-9, characterized in that: In step S3, the specific steps of the phenol-sulfuric acid method are as follows: S31: Weigh out Dendrobium officinale powder, place it in an Erlenmeyer flask, add water, and heat under reflux; S32: After cooling, transfer to a volumetric flask, dilute to volume with water, shake well, and filter; measure the filtrate, add anhydrous ethanol, and refrigerate to precipitate; S33: After centrifugation, discard the supernatant, wash the precipitate with ethanol, and centrifuge again; dissolve the precipitate in hot water and make up to volume to obtain the test solution; S34: Measure the sample solution, add phenol solution and sulfuric acid, heat in a water bath; after cooling, measure the absorbance and calculate the polysaccharide content.

Images